MapTiler is taking beta signups for GeoSplats, a system that positions gaussian splat models directly on its web basemaps and streams them through a level of detail format the company built specifically for splats. The Swiss mapping company says its processing pipeline accepts input data of any size.
Each GeoSplat is georeferenced via GPS or manual placement and rendered together with MapTiler's 3D terrain and any of its basemaps, including custom styles from the company's map editor. A JavaScript SDK is already documented, with API references and examples covering splat model loading, mouse events, basemap switching, and per-object transparency and visibility controls. A public demo is live, with the company recommending current Chrome or Safari on Windows 10/11, iOS 26+, or Android 12+.
In a February update, MapTiler reported that its lowest LOD setting represents a scene with 0.2 percent of the original data, a 3.51 million splat model rendered with roughly 17,000 splats, which is what makes streaming unbounded areas into a browser viable. That puts GeoSplats in the same problem space as the streamed SOG LOD work PlayCanvas shipped in SuperSplat and the Gaussian splat tiles Cesium has been building toward, but with MapTiler's angle being tight integration with its existing basemap, terrain, and hosting stack rather than a general purpose engine.
GeoSplats was first shown at the MapTiler Connect conference in October 2025 and remains exclusive access. The company says it is preparing a first release in the coming months and is recruiting beta testers through a signup form on the product page. MapTiler is positioning the format as AI-ready, arguing the structure suits machine learning analysis for monitoring and digital twin applications, with tourism, real estate, and construction named as target industries.
